Simulation Method and System for Path Planning of Front Loading Crane Operation in Container Yard

Abstract

The present application relates to the technical field of path analysis, and discloses a path planning simulation method and system for the operation of a reach stacker in a container yard. The method includes: performing multi-scale path topology analysis on a dynamic cost map, a density cost map, and a safety cost map to obtain potential path manifold data; performing non-stationary state quantization analysis on the potential path manifold data to obtain path dynamics characteristics, where the path dynamics characteristics include path differential geometry characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics; according to a preset path performance standard, performing multi-modal feature fusion and hierarchical weight assignment on the path dynamics characteristics to obtain an optimal path score and dynamic performance parameters for the operation of the reach stacker with spatio-temporal adaptability, and generating a self-correcting path planning scheme according to the optimal path score and dynamic performance parameters for the operation of the reach stacker. The present application improves the accuracy of path planning simulation for the operation of a reach stacker in a container yard.

Description

Technical Field

[0001] This application relates to the field of path analysis, and in particular to a path planning simulation method and system for the operation of a reach stacker in a container yard. Background Art

[0002] In the field of intelligent management of container yards, traditional path planning methods mainly rely on static path planning algorithms and simple obstacle avoidance strategies. Existing technologies usually adopt classic path planning methods such as the A* algorithm and the Dijkstra algorithm, and combine the kinematic model of the reach stacker to generate paths. As the core loading and unloading equipment in the rail-road intermodal container yard, the operation path of the reach stacker has non-linear operation characteristics, which thus constitutes complex non-linear kinematic constraints. This makes traditional path planning methods face many challenges in practical applications. At the same time, some improvement schemes introduce sampling-based fast path planning algorithms, such as the RRT (Rapidly-Exploring Random Tree) algorithm and the probabilistic roadmap method, and handle the mutual interference problem between devices by adding a simple dynamic obstacle avoidance module. These methods perform well when dealing with small-scale and low-complexity path planning tasks.

[0003] However, the existing technologies have the following deficiencies: First, traditional methods often simplify the path planning problem to the shortest path search in a static environment, ignoring the dynamic coupling effect and time-varying constraint conditions between the containers and the reach stacker in the container yard; Second, the existing obstacle avoidance strategies are mostly based on simple geometric relationship judgments and cannot effectively handle complex interaction scenarios during multi-device collaborative operations; Third, traditional methods lack the ability to systematically model and quantitatively analyze the spatio-temporal uncertainty of paths, resulting in path conflicts and efficiency losses easily occurring during actual operation; Finally, the existing technologies generally do not consider the coupling effect of equipment dynamics characteristics and environmental constraints, making the generated paths may be infeasible or unstable during execution. Summary of the Invention

[0004] This application provides a path planning simulation method and system for the operation of a reach stacker in a container yard, which is used to improve the accuracy of the optimal path planning simulation for the operation of the reach stacker in the container yard.

[0005] In a first aspect, the present application provides a path planning simulation method for the operation of a reach stacker in a container yard. The path planning simulation method for the operation of a reach stacker in a container yard includes: performing spatio-temporal correlation analysis on the container yard and the reach stacker to obtain digital parameters of the container yard; based on the digital parameters of the container yard, performing non-linear multi-body dynamics three-dimensional modeling on the preset container yard space to obtain a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data, and time-varying boundary condition data; performing adaptive resolution multi-layer grid division analysis on the heterogeneous data set to obtain a dynamic cost map considering topological changes, a density cost map based on entropy value, and a safety cost map based on collision probability; performing multi-scale path topology analysis on the dynamic cost map, the density cost map, and the safety cost map to obtain potential path manifold data with spatio-temporal uncertainty; performing non-stationary state quantization analysis on the potential path manifold data to obtain path dynamics characteristics, where the path dynamics characteristics include path differential geometry characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics; according to a preset path performance standard, performing multi-modal feature fusion and hierarchical weight assignment on the path dynamics characteristics to obtain an optimal path score and dynamic performance parameters for the operation of the reach stacker with spatio-temporal adaptability, and generating a self-correcting path planning scheme according to the optimal path score and the dynamic performance parameters for the operation of the reach stacker.

[0006] In a second aspect, the present application provides a path planning simulation system for the operation of a reach stacker in a container yard. The path planning simulation system for the operation of a reach stacker in a container yard includes:

[0007] An analysis module for performing spatio-temporal correlation analysis on the container yard and the reach stacker to obtain digital parameters of the container yard;

[0008] A modeling module for performing non-linear multi-body dynamics three-dimensional modeling on the preset container yard space based on the digital parameters of the container yard to obtain a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data, and time-varying boundary condition data;

[0009] A division module for performing adaptive resolution multi-layer grid division analysis on the heterogeneous data set to obtain a dynamic cost map considering topological changes, a density cost map based on entropy value, and a safety cost map based on collision probability;

[0010] A topology module for performing multi-scale path topology analysis on the dynamic cost map, the density cost map, and the safety cost map to obtain potential path manifold data with spatio-temporal uncertainty;

[0011] A quantization module, configured to perform non-stationary state quantization analysis on the potential path manifold data to obtain path dynamics characteristics, where the path dynamics characteristics include path differential geometry characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics;

[0012] An allocation module, configured to perform multi-modal feature fusion and hierarchical weight allocation on the path dynamics characteristics according to a preset path performance standard, to obtain an optimal path score and dynamics performance parameters for the operation of the reach stacker with spatio-temporal adaptability, and generate a self-correcting path planning scheme according to the optimal path score and the dynamics performance parameters of the reach stacker operation.

[0013] In the technical solution provided by the present application, by performing spatio-temporal correlation analysis on the container yard and the reach stacker to obtain digital parameters of the container yard, an accurate digital representation of the yard environment is realized, effectively reflecting the dynamic interaction relationship between the reach stacker and the yard environment. Based on the digital parameters of the container yard, a non-linear multi-body dynamics three-dimensional model is established, obtaining a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data, and time-varying boundary condition data, accurately describing the complex coupling relationship and dynamic constraint conditions among multiple devices. Through adaptive resolution multi-level grid division analysis of the heterogeneous data set, a dynamic cost map considering topological changes, a density cost map based on entropy value, and a safety cost map based on collision probability are generated, realizing multi-dimensional evaluation of the path planning environment. Through multi-scale path topology analysis of these cost maps, potential path manifold data with spatio-temporal uncertainty is obtained, effectively capturing the dynamic change characteristics of the path. By performing non-stationary state quantization analysis on the potential path manifold data, path dynamics characteristics including path differential geometry characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics are obtained, comprehensively characterizing the geometric and dynamic attributes of the path. Finally, based on a preset path performance standard, multi-modal feature fusion and hierarchical weight allocation are performed on the path dynamics characteristics, obtaining an optimal path score and dynamics performance parameters for the operation of the reach stacker with spatio-temporal adaptability, and generating a self-correcting path planning scheme, ensuring the real-time self-adaptability and robustness of the planning scheme. Through multi-level path analysis and feature extraction, accurate planning of the non-linear operation path of the reach stacker in a complex yard environment is realized, improving the accuracy and reliability of path planning. Description of the Drawings

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0015] Figure 1Schematic diagram of an embodiment of the path planning simulation method for the operation of a reach stacker in a container yard according to an embodiment of the present application;

[0016] Figure 2 Schematic diagram of an embodiment of the path planning simulation system for the operation of a reach stacker in a container yard according to an embodiment of the present application. Detailed implementation manners

[0017] The embodiments of the present application provide a path planning simulation method and system for the operation of a reach stacker in a container yard. Terms such as "first", "second", "third", "fourth", etc. (if any) in the specification, claims and the above drawings of the present application are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present application will be described below. Please refer to Figure 1 , an embodiment of the path planning simulation method for the operation of a reach stacker in a container yard according to an embodiment of the present application includes:

[0019] Step S101, perform spatio-temporal correlation analysis on the container yard and the reach stacker to obtain digital parameters of the container yard;

[0020] Step S102, based on the digital parameters of the container yard, perform non-linear multi-body dynamics three-dimensional modeling on the preset container yard space to obtain a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data, and time-varying boundary condition data;

[0021] Step S103, perform adaptive resolution multi-layer grid division analysis on the heterogeneous data set to obtain a dynamic cost map considering topological changes, a density cost map based on entropy value, and a safety cost map based on collision probability;

[0022] Step S104, perform multi-scale path topology analysis on the dynamic cost map, the density cost map, and the safety cost map to obtain potential path manifold data with spatio-temporal uncertainty;

[0023] Step S105: Perform non-stationary state quantization analysis on the potential path manifold data to obtain path dynamics characteristics, which include path differential geometry characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics.

[0024] Step S106: According to the preset path performance criteria, perform multi-modal feature fusion and hierarchical weight assignment on the path dynamics characteristics to obtain the optimal path score and dynamic performance parameters for the operation of the reach stacker with spatio-temporal adaptability, and generate a self-correcting path planning scheme based on the optimal path score and dynamic performance parameters of the reach stacker operation.

[0025] It can be understood that the execution subject of this application can be a path planning simulation system for the operation of the reach stacker in the container yard, and the main hardware facilities relied on are servers. This embodiment of the application takes the server as the execution subject as an example for illustration.

[0026] Specifically, perform spatio-temporal correlation analysis on the container yard and the reach stacker. Through spatial grid analysis of information such as container positions, channel distributions, and loading and unloading areas in the static layout data of the yard, and combined with the dynamic data such as the real-time position, speed, spreader height, and slewing angle of the reach stacker for time series processing, digital parameters reflecting the overall state of the yard are formed. As a device specifically used for container loading and unloading operations, the unique structural characteristics (such as the spreader system, control system, etc.) and kinematic characteristics of the reach stacker are all included in the analysis scope. Specifically, when the reach stacker moves in the yard, its position data is collected in real time and time windows are divided, and spatio-temporal characteristics are extracted through a bidirectional long short-term memory network (LSTM) to obtain accurate container yard-reach stacker interaction characteristics.

[0027] Based on the obtained digital parameters of the container yard, perform non-linear multi-body dynamics three-dimensional modeling on the preset container yard space. In this process, each reach stacker is regarded as a composite rigid body system, and motion equations including state variables such as position, speed, and acceleration are established, while considering the mutual forces and constraint conditions between the devices. In particular, the model also includes the unique dynamic parameters of the reach stacker, such as the swing effect of the spreader system, the load characteristics of the lifting mechanism, and the influence of motion inertia. By analyzing the coupling relationship between the devices, a heterogeneous data set reflecting the dynamic characteristics of the entire system is obtained.

[0028] Subsequently, an adaptive resolution multi-level grid division analysis is performed on the heterogeneous dataset, and different grid division strategies with different densities are adopted in different regions. Finer grids are used in areas with frequent operations of reach stackers (such as loading and unloading areas, empty and full container yards, etc.) to improve accuracy, and coarser grids are used in open areas to improve calculation efficiency. Based on the grid division results, a dynamic cost map, a density cost map, and a safety cost map are constructed respectively to evaluate various aspects of path planning. After obtaining the three cost maps, multi-scale path topology analysis is carried out. By extracting and analyzing path features at different scales, manifold data reflecting the spatio-temporal uncertainty of the path is obtained. These data include the geometric features, topological structure, and dynamic change characteristics of the path.

[0029] For the obtained potential path manifold data, the dynamic characteristics of the path are extracted through non-stationary state quantization analysis. These characteristics include differential geometric characteristics describing the geometric properties of the path, entropy characteristics representing the spatio-temporal distribution of the path, and gradient characteristics reflecting energy changes. According to the preset path performance criteria, the path dynamic characteristics are fused and weighted. Feature fusion is performed through a deep probability map network, and a hierarchical weight assignment strategy is adopted. Finally, the optimal path score and dynamic performance parameters for the reach stacker operation are obtained, and a self-correcting path planning scheme is generated.

[0030] For example, in a container yard, a reach stacker needs to move from point A to point B to carry a 40-foot container. Through spatio-temporal correlation analysis, digital parameters of the container yard such as the initial position (0,0), target position (50, 30) (unit: meters), and spreader height of 12 meters of the reach stacker are obtained. After non-linear multi-body dynamics three-dimensional modeling, considering the influence of two other reach stackers operating in the area, the coupling forces and constraint conditions between the three devices are calculated. At the same time, considering the structural characteristics of the reach stacker, the swing effect and load state of the spreader system are included in the modeling scope. Then, the operation area is grid-divided, with a grid density of 0.2 meters within 2 meters around the equipment and a grid density of 0.5 meters in other areas, generating a cost map reflecting all dimensions of path planning. Through multi-scale analysis and feature extraction, an optimal path considering dynamic obstacle avoidance, energy consumption, and safety is finally generated. The total length of this path is 85 meters, and the estimated running time is 300 seconds, which can ensure that the reach stacker completes the operation task safely and efficiently.

[0031] In the embodiments of the present application, by performing spatio-temporal correlation analysis on the container yard and the reach stacker, digital parameters of the container yard are obtained, realizing the accurate digital expression of the yard environment and effectively reflecting the dynamic interaction relationship between the reach stacker and the environment. Based on the digital parameters of the container yard, a non-linear multi-body dynamics three-dimensional model is established, obtaining a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data, and time-varying boundary condition data, accurately describing the complex coupling relationship and dynamic constraint conditions among multiple devices. By performing adaptive resolution multi-layer grid division analysis on the heterogeneous data set, a dynamic cost map considering topological changes, a density cost map based on entropy value, and a safety cost map based on collision probability are generated, realizing the multi-dimensional evaluation of the path planning environment. By performing multi-scale path topology analysis on these cost maps, potential path manifold data with spatio-temporal uncertainty is obtained, effectively capturing the dynamic change characteristics of the path. By performing non-stationary state quantization analysis on the potential path manifold data, path dynamics characteristics including path differential geometry characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics are obtained, comprehensively characterizing the geometric and dynamic attributes of the path. Finally, based on the preset path performance criteria, multi-modal feature fusion and hierarchical weight assignment are performed on the path dynamics characteristics, obtaining the optimal path score and dynamic performance parameters for the reach stacker operation with spatio-temporal adaptability, and generating a self-correcting path planning scheme, ensuring the real-time self-adaptability and robustness of the planning scheme. Through multi-level path analysis and feature extraction, the accurate planning of the reach stacker path in a complex yard environment is realized, improving the accuracy and reliability of path planning.

[0032] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0033] (1) Perform spatial grid analysis and processing on the static layout data of the container yard to obtain a yard basic space parameter table, and perform time series segmentation processing on the real-time position data of the reach stacker to obtain an equipment dynamic trajectory sequence;

[0034] (2) Perform topological structure analysis on the channel area data in the yard basic space parameter table to obtain a channel connectivity map, and perform time window segmentation processing on the equipment dynamic trajectory sequence to obtain an equipment behavior feature sequence;

[0035] (3) Calculate the node importance of the channel connectivity map to obtain a set of key path nodes, and perform time-frequency domain conversion processing on the equipment behavior feature sequence to obtain an equipment typical working condition spectrum;

[0036] (4) Extract spatio-temporal features from the set of critical path nodes and the typical operating conditions spectrum of the equipment through a bidirectional long short-term memory network, obtain the interaction feature matrix of the container yard - reach stacker, and perform principal component dimensionality reduction on the interaction feature matrix of the container yard - reach stacker to obtain the critical spatio-temporal feature vector;

[0037] (5) Normalize the critical spatio-temporal feature vector to obtain a standardized feature dataset, and perform multi-modal separation on the standardized feature dataset through tensor decomposition to obtain the digital parameters of the container yard.

[0038] Specifically, perform spatial grid analysis on the static layout data of the yard, divide the container yard space into multiple grid cells, and each grid cell contains information such as position coordinates and occupancy status. Calculate the spatial parameter values of each grid through spatial grid analysis:

[0039]

[0040] Among them represents the spatial parameter value of grid represents the grid occupancy, represents the distance to the nearest obstacle, represents the grid weight coefficient. At the same time, divide the real-time position data of the reach stacker at fixed time intervals Δt to generate the equipment dynamic trajectory sequence.

[0041] When performing topological structure analysis on the channel area data in the yard basic spatial parameter table, use the following connectivity calculation formula:

[0042]

[0043] Among them represents the connectivity value of node v, represents the set of nodes adjacent to node represents the reachability coefficient between nodes u and v, represents the distance between nodes. After time window slicing of the equipment dynamic trajectory sequence, extract the behavior characteristics within each time window, including parameters such as speed, acceleration, and steering angle.

[0044] Obtain the critical path nodes by calculating the node importance:

[0045]

[0046] Among them represents the importance of node represents the forward connectivity, Indicates backward connectivity, Indicates the historical access frequency of nodes, , , are weight coefficients. Perform a fast Fourier transform on the device behavior feature sequence to obtain frequency domain features and form a typical working condition spectrum of the device. Feature extraction is performed through a bidirectional long short-term memory network, which contains two LSTM layers, a forward layer and a backward layer. Each LSTM unit contains an input gate, a forget gate, and an output gate, and performs spatio-temporal feature fusion on key path nodes and the working condition spectrum. Then, the principal component analysis method is used to reduce the dimension of the high-dimensional feature matrix, and the principal components with a cumulative contribution rate reaching 95% are retained. Finally, the standardized feature data set is decomposed into multiple modal components through tensor decomposition.

[0047] For example: A container yard has an area of 100×80 meters and is divided into 50×40 grid cells. Analyze the reach stacker numbered E001 and collect its position data within 24 hours at an interval of 1 second. Through spatial grid analysis, the spatial parameter value of grid (25, 20) is calculated as P(25, 20)=0.85, indicating that this position is suitable as a channel node. For the channel node perform connectivity analysis, and the reachability coefficients of its adjacent nodes , are 0.9 and 0.8 respectively, and the distances are 2 meters and 3 meters. The connectivity value C(v1)=0.73 is calculated. In the calculation of node importance, set =0.4, =0.3, =0.3, and calculate the importance I( )=0.82 for the node . Finally, a 300-dimensional digital parameter vector of the container yard is obtained through path analysis, providing basic data support for subsequent path planning.

[0048] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0049] (1) Perform non-linear dynamics three-dimensional modeling on the equipment parameters in the digital parameters of the container yard to obtain the equipment motion state equation, and perform multi-body system coupling analysis on the equipment motion state equation to obtain equipment coupling state data;

[0050] (2) Perform constraint condition analysis on the equipment coupling state data to obtain an initial constraint condition set, and perform dynamic propagation link analysis on the initial constraint condition set to obtain dynamic constraint propagation data;

[0051] (3) Extract the boundary features of the dynamically constrained propagation data to obtain a fixed boundary constraint set, and perform time-varying perturbation analysis on the fixed boundary constraint set to obtain time-varying boundary condition data;

[0052] (4) Perform relative motion analysis on the equipment position data in the equipment coupling state data to obtain the force data between equipment, and perform state space reconstruction on the force data between equipment to obtain a coupled dynamics equation set;

[0053] (5) Perform heterogeneous feature combination processing on the equipment coupling state data, dynamically constrained propagation data, and time-varying boundary condition data through a multimodal data fusion algorithm to obtain a heterogeneous data set.

[0054] Specifically, perform a three-dimensional non-linear dynamics modeling on the equipment parameters in the digital parameters of the container yard. Consider the reachstacker as a rigid body with mass and moment of inertia, and establish the equipment motion state equation:

[0055]

[0056] Where represents the equipment state vector, including position and attitude, M is the mass matrix, is the Jacobian matrix, is the force vector, is the generalized force. For the coupled system of multiple pieces of equipment, use the constraint equation:

[0057]

[0058] Where represents the constraint function, to represent the state vectors of n pieces of equipment. Thus, the equipment coupling state data is obtained, which contains the position, velocity, and acceleration information of each piece of equipment. Then, perform constraint condition analysis on the equipment coupling state data to construct a constraint propagation graph:

[0059]

[0060] Where is the constraint propagation intensity, is the coupling coefficient between equipment i and j, It is a constraint influence factor. By solving this equation, dynamic constraint propagation data is obtained, which describes the transfer law of constraint conditions between devices. Extract the boundary features from the dynamic constraint propagation data to form a fixed boundary constraint set, and then analyze the influence of time-varying disturbances to obtain time-varying boundary condition data. Conduct relative motion analysis on the position information in the device coupling state data, calculate the acting forces between devices, and construct a coupled dynamics equation set. Finally, through a multi-modal data fusion algorithm, combine the features of various types of data to generate a heterogeneous data set.

[0061] For example: There are 3 reach stackers operating simultaneously in a container yard, numbered E001, E002, and E003 respectively, with a mass of 20 tons each and a moment of inertia of 15000 kg·m². Through non-linear dynamics three-dimensional modeling, the mass matrix M = [20000, 0; 0, 15000] and the acting force vector F = [5000N, 2000N] in the motion state equation of device E001 are obtained. Analyze the constraint relationship between E001 and E002, and the constraint propagation intensity G(t) = 0.85 is obtained, indicating a strong coupling effect between the two devices. After multi-modal data fusion, a heterogeneous data set containing position, velocity, constraints, and boundary conditions is finally obtained, and the data dimension is 500×8.

[0062] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0063] (1) Perform multi-layer grid density calculation processing on the spatio-temporal state data in the heterogeneous data set to obtain a basic grid division matrix, and perform topological structure change analysis processing on the basic grid division matrix to obtain a dynamic cost map;

[0064] (2) Perform entropy value feature extraction processing on the device coupling state data in the heterogeneous data set to obtain a regional entropy value distribution matrix, and perform density mapping processing on the regional entropy value distribution matrix to obtain a density cost map;

[0065] (3) Perform collision risk prediction processing on the time-varying boundary condition data in the heterogeneous data set to obtain a collision probability distribution matrix, and perform safety level division processing on the collision probability distribution matrix to obtain a safety cost map;

[0066] (4) Perform time series segmentation processing on the topological change data in the dynamic cost map to obtain a topological change feature sequence, and perform spatial correlation analysis processing on the topological change feature sequence to obtain grid dynamic update parameters;

[0067] (5) Perform resolution calibration processing on the dynamic cost map, density cost map, and safety cost map through an adaptive grid refinement algorithm to obtain a multi-layer grid cost map group.

[0068] Specifically, first, calculate the multi-layer grid density using the density function:

[0069]

[0070] where is the density value of the grid point , is the weight of the th spatio-temporal state point, ( , ) are the coordinates of the state point, and is the smoothing factor. Analyze the topological structure of the obtained basic grid division matrix, examine the connection relationship between grids, and construct a dynamic cost map.

[0071] Calculate the regional entropy value feature:

[0072]

[0073] where is the entropy value of the region , is the probability distribution of the th type of equipment state in the region , and is the number of state categories. Map the entropy value to the spatial grid through the density function to form a density cost map.

[0074] Adopt a collision risk prediction model:

[0075]

[0076] where is the collision risk value at position p at time t, φ(t) is the time weight function, is the risk contribution of the jth obstacle. Classify according to the risk value to generate a safety cost map.

[0077] Segment the topological data according to a fixed time window, and extract the topological change features within each segment. Conduct a spatial correlation analysis on the feature sequence, calculate the correlation degree between adjacent grids, and obtain the parameters required for grid dynamic update. Based on the quadtree structure, determine the grid division granularity according to the cost value of each region. Use a fine grid for high-cost regions and a sparse grid for low-cost regions to achieve a multi-resolution cost map representation.

[0078] For example: In a container yard of 120×100 meters, the space is initially divided into 60×50 basic grids. Taking the position (30, 25) as an example, there are 3 state points around it, with weights [1.2, 0.8, 1.0]. Substituting into the density function, λ(30, 25)=0.75 is calculated. For the equipment state analysis of the area (35, 40), the state probability distribution is [0.3, 0.4, 0.2, 0.1], and the entropy value μ(35, 40)=1.28 is calculated. In the collision risk analysis, at t = 10s, there are two obstacles affecting the position (45, 35), and the risk value θ(10, [45, 35]) = 0.9 is calculated, which is classified as a high-risk level. Through adaptive grid refinement, the grid size in the high-risk area is set to 0.5 meters, and in the low-risk area is set to 2 meters, finally forming a complete multi-layer grid cost map set.

[0079] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0080] (1) Perform multi-scale decomposition processing on the dynamic cost map to obtain a set of topological structures at different scales, and perform connected component analysis processing on the set of topological structures at different scales to obtain a path connectivity matrix;

[0081] (2) Perform hierarchical feature extraction processing on the density cost map to obtain density gradient field data, and perform manifold embedding processing on the density gradient field data to obtain an initial manifold structure;

[0082] (3) Perform probability field construction processing on the safety cost map to obtain a time-varying probability distribution map, and perform uncertainty propagation processing on the time-varying probability distribution map to obtain state uncertainty data;

[0083] (4) Perform feature fusion processing on the path connectivity matrix and the initial manifold structure to obtain a path candidate set, and perform spatio-temporal mapping processing on the path candidate set and the state uncertainty data to obtain potential path manifold data.

[0084] Specifically, the dynamic cost map is decomposed at multiple scales and processed using a multi-level pyramid structure. Starting from the original resolution at the bottom layer, cost maps at different scales are formed by successive downsampling, where the resolution of each layer is 1 / 2 of the next layer. At each scale level, the connectivity of adjacent regions is identified through the connected component labeling algorithm. Using the 8-neighborhood connectivity judgment criterion, regions that are adjacent and have a cost value difference less than the threshold are grouped into the same connected component. For each connected component, topological features such as its area, perimeter, and centroid position are calculated, and the adjacency relationship between different connected components is established, finally generating a path connectivity matrix. Then, hierarchical feature extraction is performed on the density cost map. Using a density-based hierarchical clustering method, density features are gradually extracted from local to global. At each level, the density gradient is calculated, including the gradient magnitude and direction, forming density gradient field data. The high-dimensional density gradient field data is mapped to a low-dimensional manifold space through the locally linear embedding algorithm, maintaining the local geometric relationship between data points, and obtaining an initial manifold structure. This manifold structure reflects the internal law of the density distribution in the storage yard.

[0085] When constructing the probability field for the safety cost map, the kernel density estimation method is used to convert the discrete safety cost values into a continuous probability distribution. The probability distribution is dynamically updated through a sliding time window to obtain a time-varying probability distribution map. On this basis, the Monte Carlo sampling method is applied to simulate the propagation process of uncertainty, generating multiple sets of state samples, and statistical analysis is performed to obtain state uncertainty data. The path connectivity matrix and the initial manifold structure are feature fused. Using the method of feature weighted combination, considering the connectivity and density distribution features comprehensively, a preliminary path candidate set is generated. The path candidate set and the state uncertainty data are spatiotemporally mapped to establish the corresponding relationship between path points and state uncertainty, forming the final potential path manifold data.

[0086] For example: In a 100×100-meter container storage yard, the original resolution of the dynamic cost map is 0.5 meters per pixel. After three-level pyramid decomposition, scale layers of 1 meter per pixel and 2 meters per pixel are obtained. Taking the second layer as an example, 5 main connected components are identified, with areas of 150, 200, 180, 160, and 190 square meters respectively, and the adjacency relationship matrix between these connected components is calculated. In the density feature extraction, the density gradient field of the 200-square-meter area is analyzed, and 3 main density aggregation centers are extracted. In the safety analysis, through 1000 times of Monte Carlo sampling, the state uncertainty distribution of each region is obtained. Finally, 15 potential paths are generated through feature fusion, including 8 main paths and 7 alternative paths.

[0087] In a specific embodiment, the process of performing step S105 may specifically include the following steps:

[0088] (1) Perform differential geometric calculation and processing on the potential path manifold data to obtain a curvature change sequence, and perform feature extraction processing on the curvature change sequence to obtain path differential geometric features;

[0089] (2) Perform spatio-temporal state analysis and processing on the potential path manifold data to obtain a state transition matrix, and perform entropy value calculation and processing on the state transition matrix to obtain spatio-temporal entropy features;

[0090] (3) Perform energy function construction and processing on the potential path manifold data to obtain an energy distribution map, and perform gradient calculation and processing on the energy distribution map to obtain energy gradient features;

[0091] (4) Perform feature combination processing on the path differential geometric features, spatio-temporal entropy features, and energy gradient features to obtain path dynamics features.

[0092] Specifically, the potential path manifold data refers to the set of possible paths generated during path planning, and each path contains information such as position, direction, and speed. When performing differential geometric calculations on these data, it is first necessary to parameterize the path, and then calculate the curvature value at each path point. The curvature calculation uses the formula:

[0093]

[0094] where is the curvature of the path at point t, is the path parametric equation, and are the first and second derivatives respectively. The curvature here reflects the degree of bending of the path, and the larger the curvature value, the sharper the turn. Serialize the curvature values on the entire path to obtain a curvature change sequence. Then perform feature extraction on this sequence, including calculating the statistical features (mean, variance, peak value) of the curvature, geometric features (turning radius, arc length), and dynamic features (curvature change rate), and finally form path differential geometric features. Spatio-temporal state analysis and processing is to study the motion state of path points and their conversion rules. First, it is necessary to discretize the continuous path motion into a finite number of states, such as dividing it into basic states such as straight-line motion, turning motion, acceleration, and deceleration. The state transition probability calculation uses:

[0095]

[0096] where is the state transition probability, the probability of transitioning from state x to state y, is the transfer count value, and e is the total number of states. By statistically analyzing the state transitions at adjacent moments on the path, a state transition matrix is constructed. The entropy value of this matrix is calculated to obtain the spatio-temporal entropy feature reflecting the uncertainty of state transitions. The construction of the energy function is to analyze the energy consumption characteristics of the path. Various energy forms during the movement of the reach stacker need to be considered, including kinetic energy (related to speed), potential energy (related to height), constraint energy (related to turning), etc. By calculating the energy values at each point on the path, an energy distribution map is formed. The spatial gradient of this map is calculated to obtain the energy gradient feature reflecting the energy change trend.

[0097] The feature combination process is to fuse the above three types of features. This step needs to consider the dimensions and importance of different features. Through feature normalization and weighted combination, a complete path dynamics feature vector is formed. This feature vector comprehensively describes the geometric characteristics, state change characteristics, and energy consumption characteristics of the path.

[0098] For example: In a 120×100-meter container yard, a reach stacker loading a 15-ton container needs to move from the starting point (10,15) to the ending point (90,75). Analyze a candidate path. First, sample a point every 0.5 meters on the path, and a total of 200 path points are obtained. At the path point (45,40), the calculated curvature value is 0.156, indicating that the turn is relatively gentle at this point. Feature extraction is performed on the curvature values of the 5 sampling points before and after this point, and the local curvature mean is 0.145 and the variance is 0.012. In the state analysis, the motion states of the path points are divided into 5 categories, and the statistical transition probability ρ from straight-line motion to left turn is 0.25, and the state entropy value is 1.82. The energy analysis shows that the total energy value at this point is 48.5 kJ, and the energy gradient value is 0.65 kJ / m. Finally, these features are combined into a feature vector containing 300 components, including 100-dimensional geometric features, 100-dimensional state features, and 100-dimensional energy features.

[0099] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0100] (1) Perform multi-modal hierarchical decomposition processing on the path dynamics features to obtain a sequence of feature subspaces, and perform cross-modal self-attention calculation processing on the sequence of feature subspaces to obtain an inter-modal correlation matrix;

[0101] (2) Perform non-linear feature decoupling processing on the inter-modal correlation matrix to obtain a set of key modal components, and perform hierarchical encoding processing on the set of key modal components to obtain a multi-layer feature representation vector;

[0102] (3) Perform confidence inference processing on the multi-layer feature representation vectors through a deep probability map network to obtain a modal fusion weight matrix, and perform dynamic threshold adjustment processing on the modal fusion weight matrix to obtain an adaptive weight allocation scheme;

[0103] (4) Perform spatio-temporal consistency constraint processing on the adaptive weight allocation scheme to obtain the optimal path score for the operation of the reach stacker, and perform dynamic parameter mapping processing on the optimal path score for the operation of the reach stacker to obtain dynamic performance parameters;

[0104] (5) Perform temporal prediction processing on the optimal path score and dynamic performance parameters for the operation of the reach stacker through a bidirectional recurrent neural network to obtain a path evolution sequence, and perform feedback correction processing on the path evolution sequence to obtain an initial planning scheme;

[0105] (6) Perform robustness analysis processing on the initial planning scheme to obtain a scheme stability index, and perform policy optimization processing on the scheme stability index through a reinforcement learning algorithm to obtain an alternative scheme set;

[0106] (7) Perform multi-objective trade-off processing on the alternative scheme set to obtain a candidate planning scheme, and perform constraint verification processing on the candidate planning scheme to obtain a planning feasibility assessment;

[0107] (8) Perform uncertainty quantification processing on the planning feasibility assessment through a variational inference network to obtain a risk assessment matrix, and perform threshold screening processing on the risk assessment matrix to obtain a self-correcting path planning scheme.

[0108] Specifically, perform multi-modal hierarchical decomposition processing on the path dynamics characteristics. Multi-modal hierarchical decomposition means that the path feature data containing geometric features, dynamic features, and energy features are processed in layers according to different feature modalities. This process decomposes the original features into multiple subspaces, and each subspace represents a specific feature modality. Then, calculate the correlation relationship between different feature modalities through a cross-modal self-attention mechanism to generate an inter-modal correlation matrix. The self-attention mechanism reflects the degree of mutual influence between different features by calculating the similarity scores between different modal features. Perform non-linear feature decoupling on the obtained inter-modal correlation matrix, aiming to separate the key components in each modality. The decoupling process uses non-linear transformation to separate the mutually coupled features and obtains a set of independent key modal components. These components are hierarchically encoded to form multi-layer feature representation vectors. The hierarchical encoding adopts a multi-layer encoding structure, gradually extracting more abstract feature representations from low-level local features to high-level global features.

[0109] Next, a deep probabilistic graph network is used to perform confidence inference on the multi-layer feature representation vectors. By establishing the probabilistic dependence relationships between feature nodes, the deep probabilistic graph network evaluates the reliability of different features and generates a modal fusion weight matrix. Dynamic threshold adjustment is performed on this weight matrix to adaptively adjust the weights of each modality according to the real-time state, forming an adaptive weight allocation scheme. Spatiotemporal consistency constraints are imposed on the weight allocation scheme to ensure the continuity of the path score in both time and space, obtaining the optimal path score for the operation of the reach stacker. The path score is mapped to the specific physical parameter space through a dynamic model to obtain the dynamic performance parameters including parameters such as speed and acceleration. Then, a bidirectional recurrent neural network is used to perform temporal prediction on these parameters, generating the evolution sequence of the path, and the prediction result is corrected through a feedback mechanism to form an initial planning scheme.

[0110] Robustness analysis is performed on the initial scheme to evaluate the stability of the scheme under different working conditions. The decision-making strategy of the scheme is optimized through a reinforcement learning algorithm to generate multiple alternative schemes. Multi-objective trade-offs are made on these alternative schemes, considering multiple objectives such as efficiency and safety, and candidate planning schemes are screened out. Constraint verification is performed on the candidate schemes to evaluate their feasibility. Finally, a variational inference network is used to quantify the uncertainty of the planning scheme, generating a risk assessment matrix. The schemes are screened by setting a risk threshold, and finally a self-correcting path planning scheme is obtained.

[0111] For example: In a 120×100-meter container yard, path planning is performed for a device loading a 15-ton container. The original 300-dimensional path dynamics features are first decomposed into 5 subspaces, each subspace containing 60-dimensional features. A 5×5 modal correlation matrix is obtained through self-attention calculation, where the correlation strength between geometric features and dynamics features is 0.85. After non-linear decoupling and hierarchical encoding, a 150-dimensional multi-layer feature representation vector is obtained. After analysis by the deep probabilistic graph network, the fusion weights of each modality are obtained, where the weight of the dynamics feature is 0.4, the weight of the geometric feature is 0.35, and the weight of the energy feature is 0.25. The path evolution prediction generates a state sequence of 10 time steps, and each state contains parameters such as position, speed, and acceleration. Through robustness analysis and multi-objective trade-offs, 10 candidate schemes are screened out from 50 alternative schemes. The final self-correcting path planning scheme has characteristic parameters of a total length of 108 meters, an average speed of 2.5 m / s, and a maximum steering angle of 30 degrees, meeting the requirements of safety and efficiency.

[0112] The path planning simulation method for the operation of the reach stacker in the container yard in the embodiment of the present application is described above. Next, the path planning simulation system for the operation of the reach stacker in the embodiment of the present application will be described. Please refer to Figure 2, an embodiment of the path planning simulation system for the operation of a reachstacker in a container yard in an embodiment of the present application includes:

[0113] An analysis module 201, configured to perform spatio-temporal correlation analysis on the container yard and the reachstacker to obtain digital parameters of the container yard;

[0114] A modeling module 202, configured to perform non-linear multi-body dynamics three-dimensional modeling on a preset container yard space based on the digital parameters of the container yard to obtain a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data, and time-varying boundary condition data;

[0115] A partitioning module 203, configured to perform adaptive resolution multi-layer grid partitioning analysis on the heterogeneous data set to obtain a dynamic cost map considering topological changes, a density cost map based on entropy value, and a safety cost map based on collision probability;

[0116] A topology module 204, configured to perform multi-scale path topology analysis on the dynamic cost map, the density cost map, and the safety cost map to obtain potential path manifold data with spatio-temporal uncertainty;

[0117] A quantization module 205, configured to perform non-stationary state quantization analysis on the potential path manifold data to obtain path dynamics characteristics, where the path dynamics characteristics include path differential geometric characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics;

[0118] An allocation module 206, configured to perform multi-modal feature fusion and hierarchical weight allocation on the path dynamics characteristics according to a preset path performance standard to obtain an optimal path score and dynamic performance parameters for the operation of the reachstacker with spatio-temporal adaptability, and generate a self-correcting path planning scheme according to the optimal path score and the dynamic performance parameters for the operation of the reachstacker.

[0119] Through the collaborative cooperation of the above-mentioned various components, digital parameters of the container yard are obtained by analyzing the spatio-temporal correlation between the container yard and the reach stacker, achieving an accurate digital representation of the yard environment and effectively reflecting the dynamic interaction relationship between the reach stacker and the environment. Based on the digital parameters of the container yard, a three-dimensional non-linear multi-body dynamics model is established, obtaining a heterogeneous data set containing equipment coupling state data, dynamic constraint propagation data, and time-varying boundary condition data, accurately describing the complex coupling relationship and dynamic constraint conditions among multiple devices. Through the analysis of adaptive resolution multi-level grid division of the heterogeneous data set, a dynamic cost map considering topological changes, a density cost map based on entropy value, and a safety cost map based on collision probability are generated, realizing a multi-dimensional evaluation of the path planning environment. Through multi-scale path topology analysis of these cost maps, potential path manifold data with spatio-temporal uncertainty is obtained, effectively capturing the dynamic change characteristics of the path. Through non-stationary state quantization analysis of the potential path manifold data, path dynamics characteristics including path differential geometry characteristics, spatio-temporal entropy characteristics, and energy gradient characteristics are obtained, comprehensively characterizing the geometric and dynamic attributes of the path. Finally, based on the preset path performance criteria, multi-modal feature fusion and hierarchical weight assignment are performed on the path dynamics characteristics, obtaining the optimal path score and dynamic performance parameters for the reach stacker operation with spatio-temporal adaptability, and generating a self-correcting path planning scheme, ensuring the real-time self-adaptability and robustness of the planning scheme. Through multi-level path analysis and feature extraction, accurate planning of the reach stacker path in a complex yard environment is achieved, improving the accuracy and reliability of path planning.

[0120] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A path planning simulation method for container yard reach stacker operation, characterized in that: The path planning simulation method for the container yard reach stacker operation includes: Conduct time-space correlation analysis on container yard and reach stacker to obtain digital parameters of container yard; Based on the digital parameters of the container yard, a nonlinear multi-body dynamic three-dimensional modeling is performed on the preset container yard space to obtain a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data and time-varying boundary condition data; The heterogeneous data set is subjected to adaptive resolution multi-layer grid division analysis to obtain a dynamic cost map that takes into account topological changes, a density cost map based on entropy values, and a safety cost map based on collision probability, including: performing multi-layer grid density calculation processing on the spatiotemporal state data in the heterogeneous data set to obtain a basic grid division matrix, and performing topological structure change analysis processing on the basic grid division matrix to obtain the dynamic cost map; performing entropy feature extraction processing on the device coupling state data in the heterogeneous data set to obtain a regional entropy distribution matrix, and performing density mapping processing on the regional entropy distribution matrix to obtain the density cost map; performing collision risk prediction processing on the time-varying boundary condition data in the heterogeneous data set to obtain a collision probability distribution matrix, and performing safety level division processing on the collision probability distribution matrix to obtain the safety cost map; Performing multi-scale path topology analysis on the dynamic cost map, the density cost map, and the safety cost map to obtain potential path manifold data with spatiotemporal uncertainty, where the potential path manifold data refers to a set of possible paths generated during the path planning process; Performing non-stationary state quantitative analysis on the potential path manifold data to obtain path dynamics characteristics, wherein the path dynamics characteristics include path differential geometry characteristics, spatiotemporal entropy characteristics, and energy gradient characteristics; According to the preset path performance standard, the path dynamic characteristics are subjected to multimodal feature fusion and hierarchical weight allocation to obtain the optimal path score and dynamic performance parameters of the reach loader operation with temporal and spatial adaptability, and a self-correcting path planning scheme is generated based on the optimal path score and the dynamic performance parameters of the reach loader operation.

2. The path planning simulation method for container yard reach stacker operation according to claim 1, characterized in that: The spatial and temporal correlation analysis of the container yard and the reach stacker is performed to obtain digital parameters of the container yard, including: Perform spatial grid analysis on the static layout data of the container yard to obtain the basic spatial parameter table of the yard, and perform time series segmentation on the real-time position data of the reach stacker to obtain the dynamic trajectory sequence of the equipment; Performing a topological structure analysis on the channel area data in the basic spatial parameter table of the storage yard to obtain a channel connectivity map, and performing time window segmentation processing on the equipment dynamic trajectory sequence to obtain an equipment behavior feature sequence; Calculating the node importance of the channel connectivity graph to obtain a critical path node set, and performing time-frequency domain conversion processing on the device behavior feature sequence to obtain a typical operating condition spectrum of the device; The key path node set and the typical operating condition spectrum of the equipment are extracted through a bidirectional long short-term memory network to obtain a yard-equipment interaction feature matrix, and the yard-equipment interaction feature matrix is ​​subjected to principal component dimensionality reduction processing to obtain a key time-space feature vector; The key spatiotemporal feature vectors are normalized to obtain a standardized feature data set, and the standardized feature data set is subjected to multimodal separation processing through tensor decomposition to obtain digital parameters of the container yard.

3. The path planning simulation method for container yard reach stacker operation according to claim 1, characterized in that: Based on the digital parameters of the container yard, nonlinear multi-body dynamics three-dimensional modeling is performed on the preset container yard space to obtain a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data and time-varying boundary condition data, including: Performing nonlinear dynamic three-dimensional modeling processing on the equipment parameters in the digital parameters of the container yard to obtain the equipment motion state equation, and performing multi-body system coupling analysis processing on the equipment motion state equation to obtain the equipment coupling state data; Performing constraint condition analysis processing on the device coupling state data to obtain an initial constraint condition set, and performing dynamic propagation link analysis processing on the initial constraint condition set to obtain the dynamic constraint propagation data; Performing boundary feature extraction processing on the dynamic constraint propagation data to obtain a fixed boundary constraint set, and performing time-varying disturbance analysis processing on the fixed boundary constraint set to obtain the time-varying boundary condition data; Performing relative motion analysis processing on the device position data in the device coupling state data to obtain inter-device force data, and performing state space reconstruction processing on the inter-device force data to obtain a coupling dynamics equation group; The heterogeneous data set is obtained by performing heterogeneous feature combination processing on the device coupling state data, the dynamic constraint propagation data and the time-varying boundary condition data through a multimodal data fusion algorithm.

4. The path planning simulation method for container yard reach stacker operation according to claim 1, characterized in that: The adaptive resolution multi-layer grid division analysis of the heterogeneous data set is performed to obtain a dynamic cost map considering topological changes, a density cost map based on entropy values, and a safety cost map based on collision probability, including: Performing time-series segmentation processing on the topology change data in the dynamic cost map to obtain a topology change feature sequence, and performing spatial correlation analysis processing on the topology change feature sequence to obtain a grid dynamic update parameter; The dynamic cost map, the density cost map and the safety cost map are subjected to resolution calibration processing by using an adaptive grid refinement algorithm to obtain a multi-layer grid cost map group.

5. The path planning simulation method for container yard reach stacker operation according to claim 1, characterized in that: The multi-scale path topology analysis is performed on the dynamic cost map, the density cost map, and the safety cost map to obtain potential path manifold data with spatiotemporal uncertainty, including: Performing multi-scale decomposition processing on the dynamic cost map to obtain topological structure sets of different scales, and performing connected domain analysis processing on the topological structure sets of different scales to obtain a path connectivity matrix; Performing hierarchical feature extraction processing on the density cost map to obtain density gradient field data, and performing manifold embedding processing on the density gradient field data to obtain an initial manifold structure; Performing probability field construction processing on the safety cost map to obtain a time-varying probability distribution map, and performing uncertainty propagation processing on the time-varying probability distribution map to obtain state uncertainty data; The path connectivity matrix and the initial manifold structure are subjected to feature fusion processing to obtain a path candidate set, and the path candidate set and the state uncertainty data are subjected to spatiotemporal mapping processing to obtain the potential path manifold data.

6. The path planning simulation method for container yard reach stacker operation according to claim 1, characterized in that: The non-stationary state quantitative analysis of the potential path manifold data is performed to obtain path dynamics characteristics, and the path dynamics characteristics include path differential geometry characteristics, spatiotemporal entropy characteristics and energy gradient characteristics, including: Performing differential geometry calculation processing on the potential path manifold data to obtain a curvature change sequence, and performing feature extraction processing on the curvature change sequence to obtain the path differential geometry feature; Performing spatiotemporal state analysis processing on the potential path manifold data to obtain a state transfer matrix, and performing entropy value calculation processing on the state transfer matrix to obtain the spatiotemporal entropy feature; Performing energy function construction processing on the potential path manifold data to obtain an energy distribution spectrum, and performing gradient calculation processing on the energy distribution spectrum to obtain the energy gradient feature; The path differential geometry feature, the spatiotemporal entropy feature and the energy gradient feature are subjected to feature combination processing to obtain the path dynamics feature.

7. The path planning simulation method for container yard reach stacker operation according to claim 1, characterized in that: According to the preset path performance standard, the path dynamic characteristics are subjected to multimodal feature fusion and hierarchical weight allocation to obtain the optimal path score and dynamic performance parameters of the reach stacker operation with spatiotemporal adaptability, and a self-correcting path planning scheme is generated according to the optimal path score and the dynamic performance parameters of the reach stacker operation, including: Performing multimodal hierarchical decomposition processing on the path dynamics features to obtain a feature subspace sequence, and performing cross-modal self-attention calculation processing on the feature subspace sequence to obtain an inter-modal association matrix; Performing nonlinear feature decoupling processing on the inter-modal correlation matrix to obtain a key modal component set, and performing hierarchical encoding processing on the key modal component set to obtain a multi-layer feature representation vector; Performing confidence reasoning processing on the multi-layer feature representation vector through a deep probability graph network to obtain a modal fusion weight matrix, and performing dynamic threshold adjustment processing on the modal fusion weight matrix to obtain an adaptive weight allocation scheme; Performing spatiotemporal consistency constraint processing on the adaptive weight allocation scheme to obtain an optimal path score for the reach stacker operation, and performing dynamic parameter mapping processing on the optimal path score for the reach stacker operation to obtain the dynamic performance parameter; Performing time series prediction processing on the optimal path score of the reach stacker operation and the dynamic performance parameters through a bidirectional recurrent neural network to obtain a path evolution sequence, and performing feedback correction processing on the path evolution sequence to obtain an initial planning scheme; Performing robustness analysis on the initial planning scheme to obtain a scheme stability index, and performing strategy optimization on the scheme stability index through a reinforcement learning algorithm to obtain a set of alternative schemes; Performing multi-objective trade-off processing on the set of alternative solutions to obtain candidate planning solutions, and performing constraint verification processing on the candidate planning solutions to obtain planning feasibility evaluation; The uncertainty of the planning feasibility assessment is quantified through a variational inference network to obtain a risk assessment matrix, and the risk assessment matrix is ​​subjected to threshold screening to obtain the self-correcting path planning solution.

8. A path planning simulation system for container yard reach stacker operation, used to implement the path planning simulation method for container yard reach stacker operation as claimed in any one of claims 1 to 7, characterized in that: The path planning simulation system for container yard reachstacker operation includes: The analysis module is used to perform spatiotemporal correlation analysis between the container yard and the reach stacker to obtain digital parameters of the container yard; A modeling module, for performing nonlinear multi-body dynamics three-dimensional modeling of a preset container yard space based on the digital parameters of the container yard, and obtaining a heterogeneous data set including equipment coupling state data, dynamic constraint propagation data and time-varying boundary condition data; A partitioning module is used to perform adaptive resolution multi-layer grid partitioning analysis on the heterogeneous data set to obtain a dynamic cost map that takes into account topological changes, a density cost map based on entropy values, and a safety cost map based on collision probability, including: performing multi-layer grid density calculation processing on the spatiotemporal state data in the heterogeneous data set to obtain a basic grid partitioning matrix, and performing topological structure change analysis processing on the basic grid partitioning matrix to obtain the dynamic cost map; performing entropy feature extraction processing on the device coupling state data in the heterogeneous data set to obtain a regional entropy distribution matrix, and performing density mapping processing on the regional entropy distribution matrix to obtain the density cost map; performing collision risk prediction processing on the time-varying boundary condition data in the heterogeneous data set to obtain a collision probability distribution matrix, and performing safety level partitioning processing on the collision probability distribution matrix to obtain the safety cost map; A topology module, used to perform multi-scale path topology analysis on the dynamic cost map, the density cost map and the safety cost map to obtain potential path manifold data with spatiotemporal uncertainty, where the potential path manifold data refers to a set of possible paths generated during the path planning process; A quantification module, used for performing non-stationary state quantitative analysis on the potential path manifold data to obtain path dynamics characteristics, wherein the path dynamics characteristics include path differential geometry characteristics, spatiotemporal entropy characteristics, and energy gradient characteristics; The allocation module is used to perform multimodal feature fusion and hierarchical weight allocation on the path dynamic characteristics according to a preset path performance standard, so as to obtain an optimal path score and dynamic performance parameters of a reach stacker operation with spatiotemporal adaptability, and to generate a self-correcting path planning scheme according to the optimal path score and the dynamic performance parameters of the reach stacker operation.

Images